client: Avoid memory leak in fio_client_handle_iolog()
[fio.git] / client.c
... / ...
CommitLineData
1#include <stdio.h>
2#include <stdlib.h>
3#include <unistd.h>
4#include <limits.h>
5#include <errno.h>
6#include <fcntl.h>
7#include <sys/poll.h>
8#include <sys/types.h>
9#include <sys/stat.h>
10#include <sys/wait.h>
11#include <sys/socket.h>
12#include <sys/un.h>
13#include <netinet/in.h>
14#include <arpa/inet.h>
15#include <netdb.h>
16#include <signal.h>
17#ifdef CONFIG_ZLIB
18#include <zlib.h>
19#endif
20
21#include "fio.h"
22#include "client.h"
23#include "server.h"
24#include "flist.h"
25#include "hash.h"
26#include "verify.h"
27
28static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
29static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
30static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
31static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
32static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
33static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
34static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
35
36static void convert_text(struct fio_net_cmd *cmd);
37
38struct client_ops fio_client_ops = {
39 .text = handle_text,
40 .disk_util = handle_du,
41 .thread_status = handle_ts,
42 .group_stats = handle_gs,
43 .stop = handle_stop,
44 .start = handle_start,
45 .eta = display_thread_status,
46 .probe = handle_probe,
47 .eta_msec = FIO_CLIENT_DEF_ETA_MSEC,
48 .client_type = FIO_CLIENT_TYPE_CLI,
49};
50
51static struct timespec eta_ts;
52
53static FLIST_HEAD(client_list);
54static FLIST_HEAD(eta_list);
55
56static FLIST_HEAD(arg_list);
57
58struct thread_stat client_ts;
59struct group_run_stats client_gs;
60int sum_stat_clients;
61
62static int sum_stat_nr;
63static struct json_object *root = NULL;
64static struct json_object *job_opt_object = NULL;
65static struct json_array *clients_array = NULL;
66static struct json_array *du_array = NULL;
67
68static int error_clients;
69
70#define FIO_CLIENT_HASH_BITS 7
71#define FIO_CLIENT_HASH_SZ (1 << FIO_CLIENT_HASH_BITS)
72#define FIO_CLIENT_HASH_MASK (FIO_CLIENT_HASH_SZ - 1)
73static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
74
75static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *);
76
77static void fio_client_add_hash(struct fio_client *client)
78{
79 int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
80
81 bucket &= FIO_CLIENT_HASH_MASK;
82 flist_add(&client->hash_list, &client_hash[bucket]);
83}
84
85static void fio_client_remove_hash(struct fio_client *client)
86{
87 if (!flist_empty(&client->hash_list))
88 flist_del_init(&client->hash_list);
89}
90
91static void fio_init fio_client_hash_init(void)
92{
93 int i;
94
95 for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
96 INIT_FLIST_HEAD(&client_hash[i]);
97}
98
99static int read_data(int fd, void *data, size_t size)
100{
101 ssize_t ret;
102
103 while (size) {
104 ret = read(fd, data, size);
105 if (ret < 0) {
106 if (errno == EAGAIN || errno == EINTR)
107 continue;
108 break;
109 } else if (!ret)
110 break;
111 else {
112 data += ret;
113 size -= ret;
114 }
115 }
116
117 if (size)
118 return EAGAIN;
119
120 return 0;
121}
122
123static void fio_client_json_init(void)
124{
125 char time_buf[32];
126 time_t time_p;
127
128 if (!(output_format & FIO_OUTPUT_JSON))
129 return;
130
131 time(&time_p);
132 os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
133 time_buf[strlen(time_buf) - 1] = '\0';
134
135 root = json_create_object();
136 json_object_add_value_string(root, "fio version", fio_version_string);
137 json_object_add_value_int(root, "timestamp", time_p);
138 json_object_add_value_string(root, "time", time_buf);
139
140 job_opt_object = json_create_object();
141 json_object_add_value_object(root, "global options", job_opt_object);
142 clients_array = json_create_array();
143 json_object_add_value_array(root, "client_stats", clients_array);
144 du_array = json_create_array();
145 json_object_add_value_array(root, "disk_util", du_array);
146}
147
148static void fio_client_json_fini(void)
149{
150 if (!(output_format & FIO_OUTPUT_JSON))
151 return;
152
153 log_info("\n");
154 json_print_object(root, NULL);
155 log_info("\n");
156 json_free_object(root);
157 root = NULL;
158 clients_array = NULL;
159 du_array = NULL;
160}
161
162static struct fio_client *find_client_by_fd(int fd)
163{
164 int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
165 struct fio_client *client;
166 struct flist_head *entry;
167
168 flist_for_each(entry, &client_hash[bucket]) {
169 client = flist_entry(entry, struct fio_client, hash_list);
170
171 if (client->fd == fd) {
172 client->refs++;
173 return client;
174 }
175 }
176
177 return NULL;
178}
179
180void fio_put_client(struct fio_client *client)
181{
182 if (--client->refs)
183 return;
184
185 free(client->hostname);
186 if (client->argv)
187 free(client->argv);
188 if (client->name)
189 free(client->name);
190 while (client->nr_files) {
191 struct client_file *cf = &client->files[--client->nr_files];
192
193 free(cf->file);
194 }
195 if (client->files)
196 free(client->files);
197 if (client->opt_lists)
198 free(client->opt_lists);
199
200 if (!client->did_stat)
201 sum_stat_clients--;
202
203 if (client->error)
204 error_clients++;
205
206 free(client);
207}
208
209static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
210{
211 if (!--eta->pending) {
212 eta_fn(&eta->eta);
213 free(eta);
214 return 0;
215 }
216
217 return 1;
218}
219
220static void fio_drain_client_text(struct fio_client *client)
221{
222 do {
223 struct fio_net_cmd *cmd;
224
225 cmd = fio_net_recv_cmd(client->fd, false);
226 if (!cmd)
227 break;
228
229 if (cmd->opcode == FIO_NET_CMD_TEXT) {
230 convert_text(cmd);
231 client->ops->text(client, cmd);
232 }
233
234 free(cmd);
235 } while (1);
236}
237
238static void remove_client(struct fio_client *client)
239{
240 assert(client->refs);
241
242 dprint(FD_NET, "client: removed <%s>\n", client->hostname);
243
244 fio_drain_client_text(client);
245
246 if (!flist_empty(&client->list))
247 flist_del_init(&client->list);
248
249 fio_client_remove_hash(client);
250
251 if (!flist_empty(&client->eta_list)) {
252 flist_del_init(&client->eta_list);
253 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
254 }
255
256 close(client->fd);
257 client->fd = -1;
258
259 if (client->ops->removed)
260 client->ops->removed(client);
261
262 nr_clients--;
263 fio_put_client(client);
264}
265
266struct fio_client *fio_get_client(struct fio_client *client)
267{
268 client->refs++;
269 return client;
270}
271
272static void __fio_client_add_cmd_option(struct fio_client *client,
273 const char *opt)
274{
275 int index;
276
277 index = client->argc++;
278 client->argv = realloc(client->argv, sizeof(char *) * client->argc);
279 client->argv[index] = strdup(opt);
280 dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
281}
282
283void fio_client_add_cmd_option(void *cookie, const char *opt)
284{
285 struct fio_client *client = cookie;
286 struct flist_head *entry;
287
288 if (!client || !opt)
289 return;
290
291 __fio_client_add_cmd_option(client, opt);
292
293 /*
294 * Duplicate arguments to shared client group
295 */
296 flist_for_each(entry, &arg_list) {
297 client = flist_entry(entry, struct fio_client, arg_list);
298
299 __fio_client_add_cmd_option(client, opt);
300 }
301}
302
303struct fio_client *fio_client_add_explicit(struct client_ops *ops,
304 const char *hostname, int type,
305 int port)
306{
307 struct fio_client *client;
308
309 client = malloc(sizeof(*client));
310 memset(client, 0, sizeof(*client));
311
312 INIT_FLIST_HEAD(&client->list);
313 INIT_FLIST_HEAD(&client->hash_list);
314 INIT_FLIST_HEAD(&client->arg_list);
315 INIT_FLIST_HEAD(&client->eta_list);
316 INIT_FLIST_HEAD(&client->cmd_list);
317
318 client->hostname = strdup(hostname);
319
320 if (type == Fio_client_socket)
321 client->is_sock = true;
322 else {
323 int ipv6;
324
325 ipv6 = type == Fio_client_ipv6;
326 if (fio_server_parse_host(hostname, ipv6,
327 &client->addr.sin_addr,
328 &client->addr6.sin6_addr))
329 goto err;
330
331 client->port = port;
332 }
333
334 client->fd = -1;
335 client->ops = ops;
336 client->refs = 1;
337 client->type = ops->client_type;
338
339 __fio_client_add_cmd_option(client, "fio");
340
341 flist_add(&client->list, &client_list);
342 nr_clients++;
343 dprint(FD_NET, "client: added <%s>\n", client->hostname);
344 return client;
345err:
346 free(client);
347 return NULL;
348}
349
350int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote)
351{
352 struct fio_client *client = cookie;
353 struct client_file *cf;
354 size_t new_size;
355 void *new_files;
356
357 if (!client)
358 return 1;
359
360 dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
361
362 new_size = (client->nr_files + 1) * sizeof(struct client_file);
363 new_files = realloc(client->files, new_size);
364 if (!new_files)
365 return 1;
366
367 client->files = new_files;
368 cf = &client->files[client->nr_files];
369 cf->file = strdup(ini_file);
370 cf->remote = remote;
371 client->nr_files++;
372 return 0;
373}
374
375int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
376{
377 struct fio_client *existing = *cookie;
378 struct fio_client *client;
379
380 if (existing) {
381 /*
382 * We always add our "exec" name as the option, hence 1
383 * means empty.
384 */
385 if (existing->argc == 1)
386 flist_add_tail(&existing->arg_list, &arg_list);
387 else {
388 while (!flist_empty(&arg_list))
389 flist_del_init(arg_list.next);
390 }
391 }
392
393 client = malloc(sizeof(*client));
394 memset(client, 0, sizeof(*client));
395
396 INIT_FLIST_HEAD(&client->list);
397 INIT_FLIST_HEAD(&client->hash_list);
398 INIT_FLIST_HEAD(&client->arg_list);
399 INIT_FLIST_HEAD(&client->eta_list);
400 INIT_FLIST_HEAD(&client->cmd_list);
401
402 if (fio_server_parse_string(hostname, &client->hostname,
403 &client->is_sock, &client->port,
404 &client->addr.sin_addr,
405 &client->addr6.sin6_addr,
406 &client->ipv6))
407 return -1;
408
409 client->fd = -1;
410 client->ops = ops;
411 client->refs = 1;
412 client->type = ops->client_type;
413
414 __fio_client_add_cmd_option(client, "fio");
415
416 flist_add(&client->list, &client_list);
417 nr_clients++;
418 dprint(FD_NET, "client: added <%s>\n", client->hostname);
419 *cookie = client;
420 return 0;
421}
422
423static const char *server_name(struct fio_client *client, char *buf,
424 size_t bufsize)
425{
426 const char *from;
427
428 if (client->ipv6)
429 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
430 else if (client->is_sock)
431 from = "sock";
432 else
433 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
434
435 return from;
436}
437
438static void probe_client(struct fio_client *client)
439{
440 struct cmd_client_probe_pdu pdu;
441 const char *sname;
442 uint64_t tag;
443 char buf[64];
444
445 dprint(FD_NET, "client: send probe\n");
446
447#ifdef CONFIG_ZLIB
448 pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
449#else
450 pdu.flags = 0;
451#endif
452
453 sname = server_name(client, buf, sizeof(buf));
454 memset(pdu.server, 0, sizeof(pdu.server));
455 strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
456
457 fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
458}
459
460static int fio_client_connect_ip(struct fio_client *client)
461{
462 struct sockaddr *addr;
463 socklen_t socklen;
464 int fd, domain;
465
466 if (client->ipv6) {
467 client->addr6.sin6_family = AF_INET6;
468 client->addr6.sin6_port = htons(client->port);
469 domain = AF_INET6;
470 addr = (struct sockaddr *) &client->addr6;
471 socklen = sizeof(client->addr6);
472 } else {
473 client->addr.sin_family = AF_INET;
474 client->addr.sin_port = htons(client->port);
475 domain = AF_INET;
476 addr = (struct sockaddr *) &client->addr;
477 socklen = sizeof(client->addr);
478 }
479
480 fd = socket(domain, SOCK_STREAM, 0);
481 if (fd < 0) {
482 int ret = -errno;
483
484 log_err("fio: socket: %s\n", strerror(errno));
485 return ret;
486 }
487
488 if (connect(fd, addr, socklen) < 0) {
489 int ret = -errno;
490
491 log_err("fio: connect: %s\n", strerror(errno));
492 log_err("fio: failed to connect to %s:%u\n", client->hostname,
493 client->port);
494 close(fd);
495 return ret;
496 }
497
498 return fd;
499}
500
501static int fio_client_connect_sock(struct fio_client *client)
502{
503 struct sockaddr_un *addr = &client->addr_un;
504 socklen_t len;
505 int fd;
506
507 memset(addr, 0, sizeof(*addr));
508 addr->sun_family = AF_UNIX;
509 strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1);
510
511 fd = socket(AF_UNIX, SOCK_STREAM, 0);
512 if (fd < 0) {
513 int ret = -errno;
514
515 log_err("fio: socket: %s\n", strerror(errno));
516 return ret;
517 }
518
519 len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
520 if (connect(fd, (struct sockaddr *) addr, len) < 0) {
521 int ret = -errno;
522
523 log_err("fio: connect; %s\n", strerror(errno));
524 close(fd);
525 return ret;
526 }
527
528 return fd;
529}
530
531int fio_client_connect(struct fio_client *client)
532{
533 int fd;
534
535 dprint(FD_NET, "client: connect to host %s\n", client->hostname);
536
537 if (client->is_sock)
538 fd = fio_client_connect_sock(client);
539 else
540 fd = fio_client_connect_ip(client);
541
542 dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
543
544 if (fd < 0)
545 return fd;
546
547 client->fd = fd;
548 fio_client_add_hash(client);
549 client->state = Client_connected;
550
551 probe_client(client);
552 return 0;
553}
554
555int fio_client_terminate(struct fio_client *client)
556{
557 return fio_net_send_quit(client->fd);
558}
559
560static void fio_clients_terminate(void)
561{
562 struct flist_head *entry;
563 struct fio_client *client;
564
565 dprint(FD_NET, "client: terminate clients\n");
566
567 flist_for_each(entry, &client_list) {
568 client = flist_entry(entry, struct fio_client, list);
569 fio_client_terminate(client);
570 }
571}
572
573static void sig_int(int sig)
574{
575 dprint(FD_NET, "client: got signal %d\n", sig);
576 fio_clients_terminate();
577}
578
579static void client_signal_handler(void)
580{
581 struct sigaction act;
582
583 memset(&act, 0, sizeof(act));
584 act.sa_handler = sig_int;
585 act.sa_flags = SA_RESTART;
586 sigaction(SIGINT, &act, NULL);
587
588 memset(&act, 0, sizeof(act));
589 act.sa_handler = sig_int;
590 act.sa_flags = SA_RESTART;
591 sigaction(SIGTERM, &act, NULL);
592
593/* Windows uses SIGBREAK as a quit signal from other applications */
594#ifdef WIN32
595 memset(&act, 0, sizeof(act));
596 act.sa_handler = sig_int;
597 act.sa_flags = SA_RESTART;
598 sigaction(SIGBREAK, &act, NULL);
599#endif
600
601 memset(&act, 0, sizeof(act));
602 act.sa_handler = sig_show_status;
603 act.sa_flags = SA_RESTART;
604 sigaction(SIGUSR1, &act, NULL);
605}
606
607static int send_client_cmd_line(struct fio_client *client)
608{
609 struct cmd_single_line_pdu *cslp;
610 struct cmd_line_pdu *clp;
611 unsigned long offset;
612 unsigned int *lens;
613 void *pdu;
614 size_t mem;
615 int i, ret;
616
617 dprint(FD_NET, "client: send cmdline %d\n", client->argc);
618
619 lens = malloc(client->argc * sizeof(unsigned int));
620
621 /*
622 * Find out how much mem we need
623 */
624 for (i = 0, mem = 0; i < client->argc; i++) {
625 lens[i] = strlen(client->argv[i]) + 1;
626 mem += lens[i];
627 }
628
629 /*
630 * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
631 */
632 mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
633
634 pdu = malloc(mem);
635 clp = pdu;
636 offset = sizeof(*clp);
637
638 for (i = 0; i < client->argc; i++) {
639 uint16_t arg_len = lens[i];
640
641 cslp = pdu + offset;
642 strcpy((char *) cslp->text, client->argv[i]);
643 cslp->len = cpu_to_le16(arg_len);
644 offset += sizeof(*cslp) + arg_len;
645 }
646
647 free(lens);
648 clp->lines = cpu_to_le16(client->argc);
649 clp->client_type = __cpu_to_le16(client->type);
650 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
651 free(pdu);
652 return ret;
653}
654
655int fio_clients_connect(void)
656{
657 struct fio_client *client;
658 struct flist_head *entry, *tmp;
659 int ret;
660
661#ifdef WIN32
662 WSADATA wsd;
663 WSAStartup(MAKEWORD(2, 2), &wsd);
664#endif
665
666 dprint(FD_NET, "client: connect all\n");
667
668 client_signal_handler();
669
670 flist_for_each_safe(entry, tmp, &client_list) {
671 client = flist_entry(entry, struct fio_client, list);
672
673 ret = fio_client_connect(client);
674 if (ret) {
675 remove_client(client);
676 continue;
677 }
678
679 if (client->argc > 1)
680 send_client_cmd_line(client);
681 }
682
683 return !nr_clients;
684}
685
686int fio_start_client(struct fio_client *client)
687{
688 dprint(FD_NET, "client: start %s\n", client->hostname);
689 return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
690}
691
692int fio_start_all_clients(void)
693{
694 struct fio_client *client;
695 struct flist_head *entry, *tmp;
696 int ret;
697
698 dprint(FD_NET, "client: start all\n");
699
700 fio_client_json_init();
701
702 flist_for_each_safe(entry, tmp, &client_list) {
703 client = flist_entry(entry, struct fio_client, list);
704
705 ret = fio_start_client(client);
706 if (ret) {
707 remove_client(client);
708 continue;
709 }
710 }
711
712 return flist_empty(&client_list);
713}
714
715static int __fio_client_send_remote_ini(struct fio_client *client,
716 const char *filename)
717{
718 struct cmd_load_file_pdu *pdu;
719 size_t p_size;
720 int ret;
721
722 dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
723
724 p_size = sizeof(*pdu) + strlen(filename) + 1;
725 pdu = malloc(p_size);
726 memset(pdu, 0, p_size);
727 pdu->name_len = strlen(filename);
728 strcpy((char *) pdu->file, filename);
729 pdu->client_type = cpu_to_le16((uint16_t) client->type);
730
731 client->sent_job = true;
732 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
733 free(pdu);
734 return ret;
735}
736
737/*
738 * Send file contents to server backend. We could use sendfile(), but to remain
739 * more portable lets just read/write the darn thing.
740 */
741static int __fio_client_send_local_ini(struct fio_client *client,
742 const char *filename)
743{
744 struct cmd_job_pdu *pdu;
745 size_t p_size;
746 struct stat sb;
747 char *p;
748 void *buf;
749 off_t len;
750 int fd, ret;
751
752 dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
753
754 fd = open(filename, O_RDONLY);
755 if (fd < 0) {
756 ret = -errno;
757 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
758 return ret;
759 }
760
761 if (fstat(fd, &sb) < 0) {
762 ret = -errno;
763 log_err("fio: job file stat: %s\n", strerror(errno));
764 close(fd);
765 return ret;
766 }
767
768 p_size = sb.st_size + sizeof(*pdu);
769 pdu = malloc(p_size);
770 buf = pdu->buf;
771
772 len = sb.st_size;
773 p = buf;
774 if (read_data(fd, p, len)) {
775 log_err("fio: failed reading job file %s\n", filename);
776 close(fd);
777 free(pdu);
778 return 1;
779 }
780
781 pdu->buf_len = __cpu_to_le32(sb.st_size);
782 pdu->client_type = cpu_to_le32(client->type);
783
784 client->sent_job = true;
785 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
786 free(pdu);
787 close(fd);
788 return ret;
789}
790
791int fio_client_send_ini(struct fio_client *client, const char *filename,
792 bool remote)
793{
794 int ret;
795
796 if (!remote)
797 ret = __fio_client_send_local_ini(client, filename);
798 else
799 ret = __fio_client_send_remote_ini(client, filename);
800
801 if (!ret)
802 client->sent_job = true;
803
804 return ret;
805}
806
807static int fio_client_send_cf(struct fio_client *client,
808 struct client_file *cf)
809{
810 return fio_client_send_ini(client, cf->file, cf->remote);
811}
812
813int fio_clients_send_ini(const char *filename)
814{
815 struct fio_client *client;
816 struct flist_head *entry, *tmp;
817
818 flist_for_each_safe(entry, tmp, &client_list) {
819 bool failed = false;
820
821 client = flist_entry(entry, struct fio_client, list);
822
823 if (client->nr_files) {
824 int i;
825
826 for (i = 0; i < client->nr_files; i++) {
827 struct client_file *cf;
828
829 cf = &client->files[i];
830
831 if (fio_client_send_cf(client, cf)) {
832 failed = true;
833 remove_client(client);
834 break;
835 }
836 }
837 }
838 if (client->sent_job || failed)
839 continue;
840 if (!filename || fio_client_send_ini(client, filename, 0))
841 remove_client(client);
842 }
843
844 return !nr_clients;
845}
846
847int fio_client_update_options(struct fio_client *client,
848 struct thread_options *o, uint64_t *tag)
849{
850 struct cmd_add_job_pdu pdu;
851
852 pdu.thread_number = cpu_to_le32(client->thread_number);
853 pdu.groupid = cpu_to_le32(client->groupid);
854 convert_thread_options_to_net(&pdu.top, o);
855
856 return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
857}
858
859static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
860{
861 dst->max_val = le64_to_cpu(src->max_val);
862 dst->min_val = le64_to_cpu(src->min_val);
863 dst->samples = le64_to_cpu(src->samples);
864
865 /*
866 * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
867 */
868 dst->mean.u.f = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
869 dst->S.u.f = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
870}
871
872static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
873{
874 int i, j;
875
876 dst->error = le32_to_cpu(src->error);
877 dst->thread_number = le32_to_cpu(src->thread_number);
878 dst->groupid = le32_to_cpu(src->groupid);
879 dst->pid = le32_to_cpu(src->pid);
880 dst->members = le32_to_cpu(src->members);
881 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep);
882
883 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
884 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
885 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
886 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
887 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
888 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
889 }
890
891 dst->usr_time = le64_to_cpu(src->usr_time);
892 dst->sys_time = le64_to_cpu(src->sys_time);
893 dst->ctx = le64_to_cpu(src->ctx);
894 dst->minf = le64_to_cpu(src->minf);
895 dst->majf = le64_to_cpu(src->majf);
896 dst->clat_percentiles = le32_to_cpu(src->clat_percentiles);
897 dst->lat_percentiles = le32_to_cpu(src->lat_percentiles);
898 dst->percentile_precision = le64_to_cpu(src->percentile_precision);
899
900 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
901 fio_fp64_t *fps = &src->percentile_list[i];
902 fio_fp64_t *fpd = &dst->percentile_list[i];
903
904 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
905 }
906
907 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
908 dst->io_u_map[i] = le32_to_cpu(src->io_u_map[i]);
909 dst->io_u_submit[i] = le32_to_cpu(src->io_u_submit[i]);
910 dst->io_u_complete[i] = le32_to_cpu(src->io_u_complete[i]);
911 }
912
913 for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
914 dst->io_u_lat_n[i] = le32_to_cpu(src->io_u_lat_n[i]);
915 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
916 dst->io_u_lat_u[i] = le32_to_cpu(src->io_u_lat_u[i]);
917 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
918 dst->io_u_lat_m[i] = le32_to_cpu(src->io_u_lat_m[i]);
919
920 for (i = 0; i < DDIR_RWDIR_CNT; i++)
921 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
922 dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
923
924 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
925 dst->total_io_u[i] = le64_to_cpu(src->total_io_u[i]);
926 dst->short_io_u[i] = le64_to_cpu(src->short_io_u[i]);
927 dst->drop_io_u[i] = le64_to_cpu(src->drop_io_u[i]);
928 }
929
930 dst->total_submit = le64_to_cpu(src->total_submit);
931 dst->total_complete = le64_to_cpu(src->total_complete);
932
933 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
934 dst->io_bytes[i] = le64_to_cpu(src->io_bytes[i]);
935 dst->runtime[i] = le64_to_cpu(src->runtime[i]);
936 }
937
938 dst->total_run_time = le64_to_cpu(src->total_run_time);
939 dst->continue_on_error = le16_to_cpu(src->continue_on_error);
940 dst->total_err_count = le64_to_cpu(src->total_err_count);
941 dst->first_error = le32_to_cpu(src->first_error);
942 dst->kb_base = le32_to_cpu(src->kb_base);
943 dst->unit_base = le32_to_cpu(src->unit_base);
944
945 dst->latency_depth = le32_to_cpu(src->latency_depth);
946 dst->latency_target = le64_to_cpu(src->latency_target);
947 dst->latency_window = le64_to_cpu(src->latency_window);
948 dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
949
950 dst->nr_block_infos = le64_to_cpu(src->nr_block_infos);
951 for (i = 0; i < dst->nr_block_infos; i++)
952 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
953
954 dst->ss_dur = le64_to_cpu(src->ss_dur);
955 dst->ss_state = le32_to_cpu(src->ss_state);
956 dst->ss_head = le32_to_cpu(src->ss_head);
957 dst->ss_limit.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
958 dst->ss_slope.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
959 dst->ss_deviation.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
960 dst->ss_criterion.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
961
962 if (dst->ss_state & __FIO_SS_DATA) {
963 for (i = 0; i < dst->ss_dur; i++ ) {
964 dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
965 dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
966 }
967 }
968}
969
970static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
971{
972 int i;
973
974 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
975 dst->max_run[i] = le64_to_cpu(src->max_run[i]);
976 dst->min_run[i] = le64_to_cpu(src->min_run[i]);
977 dst->max_bw[i] = le64_to_cpu(src->max_bw[i]);
978 dst->min_bw[i] = le64_to_cpu(src->min_bw[i]);
979 dst->iobytes[i] = le64_to_cpu(src->iobytes[i]);
980 dst->agg[i] = le64_to_cpu(src->agg[i]);
981 }
982
983 dst->kb_base = le32_to_cpu(src->kb_base);
984 dst->unit_base = le32_to_cpu(src->unit_base);
985 dst->groupid = le32_to_cpu(src->groupid);
986 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep);
987}
988
989static void json_object_add_client_info(struct json_object *obj,
990 struct fio_client *client)
991{
992 const char *hostname = client->hostname ? client->hostname : "";
993
994 json_object_add_value_string(obj, "hostname", hostname);
995 json_object_add_value_int(obj, "port", client->port);
996}
997
998static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
999{
1000 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1001 struct flist_head *opt_list = NULL;
1002 struct json_object *tsobj;
1003
1004 if (client->opt_lists && p->ts.thread_number <= client->jobs)
1005 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1006
1007 tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
1008 client->did_stat = true;
1009 if (tsobj) {
1010 json_object_add_client_info(tsobj, client);
1011 json_array_add_value_object(clients_array, tsobj);
1012 }
1013
1014 if (sum_stat_clients <= 1)
1015 return;
1016
1017 sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1018 sum_group_stats(&client_gs, &p->rs);
1019
1020 client_ts.members++;
1021 client_ts.thread_number = p->ts.thread_number;
1022 client_ts.groupid = p->ts.groupid;
1023 client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1024
1025 if (++sum_stat_nr == sum_stat_clients) {
1026 strcpy(client_ts.name, "All clients");
1027 tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
1028 if (tsobj) {
1029 json_object_add_client_info(tsobj, client);
1030 json_array_add_value_object(clients_array, tsobj);
1031 }
1032 }
1033}
1034
1035static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1036{
1037 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1038
1039 if (output_format & FIO_OUTPUT_NORMAL)
1040 show_group_stats(gs, NULL);
1041}
1042
1043static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1044{
1045 struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1046 struct print_option *p;
1047
1048 if (!job_opt_object)
1049 return;
1050
1051 pdu->global = le16_to_cpu(pdu->global);
1052 pdu->truncated = le16_to_cpu(pdu->truncated);
1053 pdu->groupid = le32_to_cpu(pdu->groupid);
1054
1055 p = malloc(sizeof(*p));
1056 p->name = strdup((char *) pdu->name);
1057 if (pdu->value[0] != '\0')
1058 p->value = strdup((char *) pdu->value);
1059 else
1060 p->value = NULL;
1061
1062 if (pdu->global) {
1063 const char *pos = "";
1064
1065 if (p->value)
1066 pos = p->value;
1067
1068 json_object_add_value_string(job_opt_object, p->name, pos);
1069 } else if (client->opt_lists) {
1070 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1071
1072 flist_add_tail(&p->list, opt_list);
1073 }
1074}
1075
1076static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1077{
1078 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1079 const char *buf = (const char *) pdu->buf;
1080 const char *name;
1081 int fio_unused ret;
1082
1083 name = client->name ? client->name : client->hostname;
1084
1085 if (!client->skip_newline)
1086 fprintf(f_out, "<%s> ", name);
1087 ret = fwrite(buf, pdu->buf_len, 1, f_out);
1088 fflush(f_out);
1089 client->skip_newline = strchr(buf, '\n') == NULL;
1090}
1091
1092static void convert_agg(struct disk_util_agg *agg)
1093{
1094 int i;
1095
1096 for (i = 0; i < 2; i++) {
1097 agg->ios[i] = le64_to_cpu(agg->ios[i]);
1098 agg->merges[i] = le64_to_cpu(agg->merges[i]);
1099 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1100 agg->ticks[i] = le64_to_cpu(agg->ticks[i]);
1101 }
1102
1103 agg->io_ticks = le64_to_cpu(agg->io_ticks);
1104 agg->time_in_queue = le64_to_cpu(agg->time_in_queue);
1105 agg->slavecount = le32_to_cpu(agg->slavecount);
1106 agg->max_util.u.f = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1107}
1108
1109static void convert_dus(struct disk_util_stat *dus)
1110{
1111 int i;
1112
1113 for (i = 0; i < 2; i++) {
1114 dus->s.ios[i] = le64_to_cpu(dus->s.ios[i]);
1115 dus->s.merges[i] = le64_to_cpu(dus->s.merges[i]);
1116 dus->s.sectors[i] = le64_to_cpu(dus->s.sectors[i]);
1117 dus->s.ticks[i] = le64_to_cpu(dus->s.ticks[i]);
1118 }
1119
1120 dus->s.io_ticks = le64_to_cpu(dus->s.io_ticks);
1121 dus->s.time_in_queue = le64_to_cpu(dus->s.time_in_queue);
1122 dus->s.msec = le64_to_cpu(dus->s.msec);
1123}
1124
1125static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1126{
1127 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1128
1129 if (!client->disk_stats_shown) {
1130 client->disk_stats_shown = true;
1131 log_info("\nDisk stats (read/write):\n");
1132 }
1133
1134 if (output_format & FIO_OUTPUT_JSON) {
1135 struct json_object *duobj;
1136 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1137 duobj = json_array_last_value_object(du_array);
1138 json_object_add_client_info(duobj, client);
1139 }
1140 if (output_format & FIO_OUTPUT_TERSE)
1141 print_disk_util(&du->dus, &du->agg, 1, NULL);
1142 if (output_format & FIO_OUTPUT_NORMAL)
1143 print_disk_util(&du->dus, &du->agg, 0, NULL);
1144}
1145
1146static void convert_jobs_eta(struct jobs_eta *je)
1147{
1148 int i;
1149
1150 je->nr_running = le32_to_cpu(je->nr_running);
1151 je->nr_ramp = le32_to_cpu(je->nr_ramp);
1152 je->nr_pending = le32_to_cpu(je->nr_pending);
1153 je->nr_setting_up = le32_to_cpu(je->nr_setting_up);
1154 je->files_open = le32_to_cpu(je->files_open);
1155
1156 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1157 je->m_rate[i] = le64_to_cpu(je->m_rate[i]);
1158 je->t_rate[i] = le64_to_cpu(je->t_rate[i]);
1159 je->m_iops[i] = le32_to_cpu(je->m_iops[i]);
1160 je->t_iops[i] = le32_to_cpu(je->t_iops[i]);
1161 je->rate[i] = le64_to_cpu(je->rate[i]);
1162 je->iops[i] = le32_to_cpu(je->iops[i]);
1163 }
1164
1165 je->elapsed_sec = le64_to_cpu(je->elapsed_sec);
1166 je->eta_sec = le64_to_cpu(je->eta_sec);
1167 je->nr_threads = le32_to_cpu(je->nr_threads);
1168 je->is_pow2 = le32_to_cpu(je->is_pow2);
1169 je->unit_base = le32_to_cpu(je->unit_base);
1170}
1171
1172void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1173{
1174 int i;
1175
1176 dst->nr_running += je->nr_running;
1177 dst->nr_ramp += je->nr_ramp;
1178 dst->nr_pending += je->nr_pending;
1179 dst->nr_setting_up += je->nr_setting_up;
1180 dst->files_open += je->files_open;
1181
1182 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1183 dst->m_rate[i] += je->m_rate[i];
1184 dst->t_rate[i] += je->t_rate[i];
1185 dst->m_iops[i] += je->m_iops[i];
1186 dst->t_iops[i] += je->t_iops[i];
1187 dst->rate[i] += je->rate[i];
1188 dst->iops[i] += je->iops[i];
1189 }
1190
1191 dst->elapsed_sec += je->elapsed_sec;
1192
1193 if (je->eta_sec > dst->eta_sec)
1194 dst->eta_sec = je->eta_sec;
1195
1196 dst->nr_threads += je->nr_threads;
1197
1198 /*
1199 * This wont be correct for multiple strings, but at least it
1200 * works for the basic cases.
1201 */
1202 strcpy((char *) dst->run_str, (char *) je->run_str);
1203}
1204
1205static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1206{
1207 struct fio_net_cmd_reply *reply = NULL;
1208 struct flist_head *entry;
1209
1210 flist_for_each(entry, &client->cmd_list) {
1211 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1212
1213 if (cmd->tag == (uintptr_t) reply)
1214 break;
1215
1216 reply = NULL;
1217 }
1218
1219 if (!reply) {
1220 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1221 return false;
1222 }
1223
1224 flist_del(&reply->list);
1225 cmd->tag = reply->saved_tag;
1226 free(reply);
1227 return true;
1228}
1229
1230int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1231{
1232 do {
1233 struct fio_net_cmd_reply *reply = NULL;
1234 struct flist_head *entry;
1235
1236 flist_for_each(entry, &client->cmd_list) {
1237 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1238
1239 if (tag == (uintptr_t) reply)
1240 break;
1241
1242 reply = NULL;
1243 }
1244
1245 if (!reply)
1246 break;
1247
1248 usleep(1000);
1249 } while (1);
1250
1251 return 0;
1252}
1253
1254static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1255{
1256 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1257 struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1258
1259 dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1260
1261 assert(client->eta_in_flight == eta);
1262
1263 client->eta_in_flight = NULL;
1264 flist_del_init(&client->eta_list);
1265 client->eta_timeouts = 0;
1266
1267 if (client->ops->jobs_eta)
1268 client->ops->jobs_eta(client, je);
1269
1270 fio_client_sum_jobs_eta(&eta->eta, je);
1271 fio_client_dec_jobs_eta(eta, client->ops->eta);
1272}
1273
1274static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1275 uint64_t sample_size)
1276{
1277 struct io_sample *s;
1278 int log_offset;
1279 uint64_t i, j, nr_samples;
1280 struct io_u_plat_entry *entry;
1281 unsigned int *io_u_plat;
1282
1283 int stride = 1 << hist_coarseness;
1284
1285 if (!sample_size)
1286 return;
1287
1288 s = __get_sample(samples, 0, 0);
1289 log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1290
1291 nr_samples = sample_size / __log_entry_sz(log_offset);
1292
1293 for (i = 0; i < nr_samples; i++) {
1294
1295 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1296 i * sizeof(struct io_u_plat_entry));
1297
1298 entry = s->data.plat_entry;
1299 io_u_plat = entry->io_u_plat;
1300
1301 fprintf(f, "%lu, %u, %u, ", (unsigned long) s->time,
1302 io_sample_ddir(s), s->bs);
1303 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1304 fprintf(f, "%lu, ", hist_sum(j, stride, io_u_plat, NULL));
1305 }
1306 fprintf(f, "%lu\n", (unsigned long)
1307 hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1308
1309 }
1310}
1311
1312static int fio_client_handle_iolog(struct fio_client *client,
1313 struct fio_net_cmd *cmd)
1314{
1315 struct cmd_iolog_pdu *pdu = NULL;
1316 bool store_direct;
1317 char *log_pathname = NULL;
1318 int ret = 0;
1319
1320 pdu = convert_iolog(cmd, &store_direct);
1321 if (!pdu) {
1322 log_err("fio: failed converting IO log\n");
1323 ret = 1;
1324 goto out;
1325 }
1326
1327 /* allocate buffer big enough for next sprintf() call */
1328 log_pathname = malloc(10 + strlen((char *)pdu->name) +
1329 strlen(client->hostname));
1330 if (!log_pathname) {
1331 log_err("fio: memory allocation of unique pathname failed\n");
1332 ret = -1;
1333 goto out;
1334 }
1335 /* generate a unique pathname for the log file using hostname */
1336 sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1337
1338 if (store_direct) {
1339 ssize_t ret;
1340 size_t sz;
1341 int fd;
1342
1343 fd = open((const char *) log_pathname,
1344 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1345 if (fd < 0) {
1346 log_err("fio: open log %s: %s\n",
1347 log_pathname, strerror(errno));
1348 ret = 1;
1349 goto out;
1350 }
1351
1352 sz = cmd->pdu_len - sizeof(*pdu);
1353 ret = write(fd, pdu->samples, sz);
1354 close(fd);
1355
1356 if (ret != sz) {
1357 log_err("fio: short write on compressed log\n");
1358 ret = 1;
1359 goto out;
1360 }
1361
1362 ret = 0;
1363 } else {
1364 FILE *f;
1365 f = fopen((const char *) log_pathname, "w");
1366 if (!f) {
1367 log_err("fio: fopen log %s : %s\n",
1368 log_pathname, strerror(errno));
1369 ret = 1;
1370 goto out;
1371 }
1372
1373 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1374 client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1375 pdu->nr_samples * sizeof(struct io_sample));
1376 } else {
1377 flush_samples(f, pdu->samples,
1378 pdu->nr_samples * sizeof(struct io_sample));
1379 }
1380 fclose(f);
1381 ret = 0;
1382 }
1383
1384out:
1385 if (pdu && pdu != cmd->payload)
1386 free(pdu);
1387
1388 if (log_pathname)
1389 free(log_pathname);
1390
1391 return ret;
1392}
1393
1394static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1395{
1396 struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1397 const char *os, *arch;
1398 char bit[16];
1399
1400 os = fio_get_os_string(probe->os);
1401 if (!os)
1402 os = "unknown";
1403
1404 arch = fio_get_arch_string(probe->arch);
1405 if (!arch)
1406 os = "unknown";
1407
1408 sprintf(bit, "%d-bit", probe->bpp * 8);
1409 probe->flags = le64_to_cpu(probe->flags);
1410
1411 log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1412 probe->hostname, probe->bigendian, bit, os, arch,
1413 probe->fio_version, (unsigned long) probe->flags);
1414
1415 if (!client->name)
1416 client->name = strdup((char *) probe->hostname);
1417}
1418
1419static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1420{
1421 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1422
1423 client->state = Client_started;
1424 client->jobs = le32_to_cpu(pdu->jobs);
1425 client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1426
1427 if (client->jobs) {
1428 int i;
1429
1430 if (client->opt_lists)
1431 free(client->opt_lists);
1432
1433 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1434 for (i = 0; i < client->jobs; i++)
1435 INIT_FLIST_HEAD(&client->opt_lists[i]);
1436 }
1437
1438 sum_stat_clients += client->nr_stat;
1439}
1440
1441static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
1442{
1443 if (client->error)
1444 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1445}
1446
1447static void convert_stop(struct fio_net_cmd *cmd)
1448{
1449 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1450
1451 pdu->error = le32_to_cpu(pdu->error);
1452}
1453
1454static void convert_text(struct fio_net_cmd *cmd)
1455{
1456 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1457
1458 pdu->level = le32_to_cpu(pdu->level);
1459 pdu->buf_len = le32_to_cpu(pdu->buf_len);
1460 pdu->log_sec = le64_to_cpu(pdu->log_sec);
1461 pdu->log_usec = le64_to_cpu(pdu->log_usec);
1462}
1463
1464static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1465 struct cmd_iolog_pdu *pdu)
1466{
1467#ifdef CONFIG_ZLIB
1468 struct cmd_iolog_pdu *ret;
1469 z_stream stream;
1470 uint32_t nr_samples;
1471 size_t total;
1472 char *p;
1473
1474 stream.zalloc = Z_NULL;
1475 stream.zfree = Z_NULL;
1476 stream.opaque = Z_NULL;
1477 stream.avail_in = 0;
1478 stream.next_in = Z_NULL;
1479
1480 if (inflateInit(&stream) != Z_OK)
1481 return NULL;
1482
1483 /*
1484 * Get header first, it's not compressed
1485 */
1486 nr_samples = le64_to_cpu(pdu->nr_samples);
1487
1488 if (pdu->log_type == IO_LOG_TYPE_HIST)
1489 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1490 sizeof(struct io_u_plat_entry));
1491 else
1492 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1493 ret = malloc(total + sizeof(*pdu));
1494 ret->nr_samples = nr_samples;
1495
1496 memcpy(ret, pdu, sizeof(*pdu));
1497
1498 p = (char *) ret + sizeof(*pdu);
1499
1500 stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1501 stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1502 while (stream.avail_in) {
1503 unsigned int this_chunk = 65536;
1504 unsigned int this_len;
1505 int err;
1506
1507 if (this_chunk > total)
1508 this_chunk = total;
1509
1510 stream.avail_out = this_chunk;
1511 stream.next_out = (void *)p;
1512 err = inflate(&stream, Z_NO_FLUSH);
1513 /* may be Z_OK, or Z_STREAM_END */
1514 if (err < 0) {
1515 log_err("fio: inflate error %d\n", err);
1516 free(ret);
1517 ret = NULL;
1518 goto err;
1519 }
1520
1521 this_len = this_chunk - stream.avail_out;
1522 p += this_len;
1523 total -= this_len;
1524 }
1525
1526err:
1527 inflateEnd(&stream);
1528 return ret;
1529#else
1530 return NULL;
1531#endif
1532}
1533
1534/*
1535 * This has been compressed on the server side, since it can be big.
1536 * Uncompress here.
1537 */
1538static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1539 bool *store_direct)
1540{
1541 struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1542 struct cmd_iolog_pdu *ret;
1543 uint64_t i;
1544 int compressed;
1545 void *samples;
1546
1547 *store_direct = false;
1548
1549 /*
1550 * Convert if compressed and we support it. If it's not
1551 * compressed, we need not do anything.
1552 */
1553 compressed = le32_to_cpu(pdu->compressed);
1554 if (compressed == XMIT_COMPRESSED) {
1555#ifndef CONFIG_ZLIB
1556 log_err("fio: server sent compressed data by mistake\n");
1557 return NULL;
1558#endif
1559 ret = convert_iolog_gz(cmd, pdu);
1560 if (!ret) {
1561 log_err("fio: failed decompressing log\n");
1562 return NULL;
1563 }
1564 } else if (compressed == STORE_COMPRESSED) {
1565 *store_direct = true;
1566 ret = pdu;
1567 } else
1568 ret = pdu;
1569
1570 ret->nr_samples = le64_to_cpu(ret->nr_samples);
1571 ret->thread_number = le32_to_cpu(ret->thread_number);
1572 ret->log_type = le32_to_cpu(ret->log_type);
1573 ret->compressed = le32_to_cpu(ret->compressed);
1574 ret->log_offset = le32_to_cpu(ret->log_offset);
1575 ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1576
1577 if (*store_direct)
1578 return ret;
1579
1580 samples = &ret->samples[0];
1581 for (i = 0; i < ret->nr_samples; i++) {
1582 struct io_sample *s;
1583
1584 s = __get_sample(samples, ret->log_offset, i);
1585 if (ret->log_type == IO_LOG_TYPE_HIST)
1586 s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1587
1588 s->time = le64_to_cpu(s->time);
1589 s->data.val = le64_to_cpu(s->data.val);
1590 s->__ddir = le32_to_cpu(s->__ddir);
1591 s->bs = le32_to_cpu(s->bs);
1592
1593 if (ret->log_offset) {
1594 struct io_sample_offset *so = (void *) s;
1595
1596 so->offset = le64_to_cpu(so->offset);
1597 }
1598
1599 if (ret->log_type == IO_LOG_TYPE_HIST) {
1600 s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1601 s->data.plat_entry->list.next = NULL;
1602 s->data.plat_entry->list.prev = NULL;
1603 }
1604 }
1605
1606 return ret;
1607}
1608
1609static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1610 size_t size, uint64_t tag)
1611{
1612 rep->error = cpu_to_le32(rep->error);
1613 fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1614}
1615
1616static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1617 uint64_t tag)
1618{
1619 struct cmd_sendfile_reply *rep;
1620 struct stat sb;
1621 size_t size;
1622 int fd;
1623
1624 size = sizeof(*rep);
1625 rep = malloc(size);
1626
1627 if (stat((char *)pdu->path, &sb) < 0) {
1628fail:
1629 rep->error = errno;
1630 sendfile_reply(client->fd, rep, size, tag);
1631 free(rep);
1632 return 1;
1633 }
1634
1635 size += sb.st_size;
1636 rep = realloc(rep, size);
1637 rep->size = cpu_to_le32((uint32_t) sb.st_size);
1638
1639 fd = open((char *)pdu->path, O_RDONLY);
1640 if (fd == -1 )
1641 goto fail;
1642
1643 rep->error = read_data(fd, &rep->data, sb.st_size);
1644 sendfile_reply(client->fd, rep, size, tag);
1645 free(rep);
1646 close(fd);
1647 return 0;
1648}
1649
1650int fio_handle_client(struct fio_client *client)
1651{
1652 struct client_ops *ops = client->ops;
1653 struct fio_net_cmd *cmd;
1654 int size;
1655
1656 dprint(FD_NET, "client: handle %s\n", client->hostname);
1657
1658 cmd = fio_net_recv_cmd(client->fd, true);
1659 if (!cmd)
1660 return 0;
1661
1662 dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1663 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1664
1665 switch (cmd->opcode) {
1666 case FIO_NET_CMD_QUIT:
1667 if (ops->quit)
1668 ops->quit(client, cmd);
1669 remove_client(client);
1670 break;
1671 case FIO_NET_CMD_TEXT:
1672 convert_text(cmd);
1673 ops->text(client, cmd);
1674 break;
1675 case FIO_NET_CMD_DU: {
1676 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1677
1678 convert_dus(&du->dus);
1679 convert_agg(&du->agg);
1680
1681 ops->disk_util(client, cmd);
1682 break;
1683 }
1684 case FIO_NET_CMD_TS: {
1685 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1686
1687 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1688 if (le32_to_cpu(p->ts.ss_state) & __FIO_SS_DATA) {
1689 dprint(FD_NET, "client: received steadystate ring buffers\n");
1690
1691 size = le64_to_cpu(p->ts.ss_dur);
1692 p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1693 p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1694 }
1695
1696 convert_ts(&p->ts, &p->ts);
1697 convert_gs(&p->rs, &p->rs);
1698
1699 ops->thread_status(client, cmd);
1700 break;
1701 }
1702 case FIO_NET_CMD_GS: {
1703 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1704
1705 convert_gs(gs, gs);
1706
1707 ops->group_stats(client, cmd);
1708 break;
1709 }
1710 case FIO_NET_CMD_ETA: {
1711 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1712
1713 if (!remove_reply_cmd(client, cmd))
1714 break;
1715 convert_jobs_eta(je);
1716 handle_eta(client, cmd);
1717 break;
1718 }
1719 case FIO_NET_CMD_PROBE:
1720 remove_reply_cmd(client, cmd);
1721 ops->probe(client, cmd);
1722 break;
1723 case FIO_NET_CMD_SERVER_START:
1724 client->state = Client_running;
1725 if (ops->job_start)
1726 ops->job_start(client, cmd);
1727 break;
1728 case FIO_NET_CMD_START: {
1729 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1730
1731 pdu->jobs = le32_to_cpu(pdu->jobs);
1732 ops->start(client, cmd);
1733 break;
1734 }
1735 case FIO_NET_CMD_STOP: {
1736 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1737
1738 convert_stop(cmd);
1739 client->state = Client_stopped;
1740 client->error = le32_to_cpu(pdu->error);
1741 client->signal = le32_to_cpu(pdu->signal);
1742 ops->stop(client, cmd);
1743 break;
1744 }
1745 case FIO_NET_CMD_ADD_JOB: {
1746 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1747
1748 client->thread_number = le32_to_cpu(pdu->thread_number);
1749 client->groupid = le32_to_cpu(pdu->groupid);
1750
1751 if (ops->add_job)
1752 ops->add_job(client, cmd);
1753 break;
1754 }
1755 case FIO_NET_CMD_IOLOG:
1756 fio_client_handle_iolog(client, cmd);
1757 break;
1758 case FIO_NET_CMD_UPDATE_JOB:
1759 ops->update_job(client, cmd);
1760 remove_reply_cmd(client, cmd);
1761 break;
1762 case FIO_NET_CMD_VTRIGGER: {
1763 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1764 char buf[128];
1765 int off = 0;
1766
1767 if (aux_path) {
1768 strcpy(buf, aux_path);
1769 off = strlen(buf);
1770 }
1771
1772 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1773 exec_trigger(trigger_cmd);
1774 break;
1775 }
1776 case FIO_NET_CMD_SENDFILE: {
1777 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1778 fio_send_file(client, pdu, cmd->tag);
1779 break;
1780 }
1781 case FIO_NET_CMD_JOB_OPT: {
1782 handle_job_opt(client, cmd);
1783 break;
1784 }
1785 default:
1786 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1787 break;
1788 }
1789
1790 free(cmd);
1791 return 1;
1792}
1793
1794int fio_clients_send_trigger(const char *cmd)
1795{
1796 struct flist_head *entry;
1797 struct fio_client *client;
1798 size_t slen;
1799
1800 dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1801
1802 if (!cmd)
1803 slen = 0;
1804 else
1805 slen = strlen(cmd);
1806
1807 flist_for_each(entry, &client_list) {
1808 struct cmd_vtrigger_pdu *pdu;
1809
1810 client = flist_entry(entry, struct fio_client, list);
1811
1812 pdu = malloc(sizeof(*pdu) + slen);
1813 pdu->len = cpu_to_le16((uint16_t) slen);
1814 if (slen)
1815 memcpy(pdu->cmd, cmd, slen);
1816 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1817 sizeof(*pdu) + slen, NULL, NULL);
1818 free(pdu);
1819 }
1820
1821 return 0;
1822}
1823
1824static void request_client_etas(struct client_ops *ops)
1825{
1826 struct fio_client *client;
1827 struct flist_head *entry;
1828 struct client_eta *eta;
1829 int skipped = 0;
1830
1831 dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1832
1833 eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1834 eta->pending = nr_clients;
1835
1836 flist_for_each(entry, &client_list) {
1837 client = flist_entry(entry, struct fio_client, list);
1838
1839 if (!flist_empty(&client->eta_list)) {
1840 skipped++;
1841 continue;
1842 }
1843 if (client->state != Client_running)
1844 continue;
1845
1846 assert(!client->eta_in_flight);
1847 flist_add_tail(&client->eta_list, &eta_list);
1848 client->eta_in_flight = eta;
1849 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1850 (uintptr_t) eta, &client->cmd_list);
1851 }
1852
1853 while (skipped--) {
1854 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1855 break;
1856 }
1857
1858 dprint(FD_NET, "client: requested eta tag %p\n", eta);
1859}
1860
1861/*
1862 * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1863 */
1864static int handle_cmd_timeout(struct fio_client *client,
1865 struct fio_net_cmd_reply *reply)
1866{
1867 flist_del(&reply->list);
1868 free(reply);
1869
1870 if (reply->opcode != FIO_NET_CMD_SEND_ETA)
1871 return 1;
1872
1873 log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1874
1875 flist_del_init(&client->eta_list);
1876 if (client->eta_in_flight) {
1877 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1878 client->eta_in_flight = NULL;
1879 }
1880
1881 /*
1882 * If we fail 5 in a row, give up...
1883 */
1884 if (client->eta_timeouts++ > 5)
1885 return 1;
1886
1887 return 0;
1888}
1889
1890static int client_check_cmd_timeout(struct fio_client *client,
1891 struct timespec *now)
1892{
1893 struct fio_net_cmd_reply *reply;
1894 struct flist_head *entry, *tmp;
1895 int ret = 0;
1896
1897 flist_for_each_safe(entry, tmp, &client->cmd_list) {
1898 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1899
1900 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
1901 continue;
1902
1903 if (!handle_cmd_timeout(client, reply))
1904 continue;
1905
1906 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1907 fio_server_op(reply->opcode));
1908 ret = 1;
1909 }
1910
1911 return flist_empty(&client->cmd_list) && ret;
1912}
1913
1914static int fio_check_clients_timed_out(void)
1915{
1916 struct fio_client *client;
1917 struct flist_head *entry, *tmp;
1918 struct timespec ts;
1919 int ret = 0;
1920
1921 fio_gettime(&ts, NULL);
1922
1923 flist_for_each_safe(entry, tmp, &client_list) {
1924 client = flist_entry(entry, struct fio_client, list);
1925
1926 if (flist_empty(&client->cmd_list))
1927 continue;
1928
1929 if (!client_check_cmd_timeout(client, &ts))
1930 continue;
1931
1932 if (client->ops->timed_out)
1933 client->ops->timed_out(client);
1934 else
1935 log_err("fio: client %s timed out\n", client->hostname);
1936
1937 client->error = ETIMEDOUT;
1938 remove_client(client);
1939 ret = 1;
1940 }
1941
1942 return ret;
1943}
1944
1945int fio_handle_clients(struct client_ops *ops)
1946{
1947 struct pollfd *pfds;
1948 int i, ret = 0, retval = 0;
1949
1950 fio_gettime(&eta_ts, NULL);
1951
1952 pfds = malloc(nr_clients * sizeof(struct pollfd));
1953
1954 init_thread_stat(&client_ts);
1955 init_group_run_stat(&client_gs);
1956
1957 while (!exit_backend && nr_clients) {
1958 struct flist_head *entry, *tmp;
1959 struct fio_client *client;
1960
1961 i = 0;
1962 flist_for_each_safe(entry, tmp, &client_list) {
1963 client = flist_entry(entry, struct fio_client, list);
1964
1965 if (!client->sent_job && !client->ops->stay_connected &&
1966 flist_empty(&client->cmd_list)) {
1967 remove_client(client);
1968 continue;
1969 }
1970
1971 pfds[i].fd = client->fd;
1972 pfds[i].events = POLLIN;
1973 i++;
1974 }
1975
1976 if (!nr_clients)
1977 break;
1978
1979 assert(i == nr_clients);
1980
1981 do {
1982 struct timespec ts;
1983 int timeout;
1984
1985 fio_gettime(&ts, NULL);
1986 if (mtime_since(&eta_ts, &ts) >= 900) {
1987 request_client_etas(ops);
1988 memcpy(&eta_ts, &ts, sizeof(ts));
1989
1990 if (fio_check_clients_timed_out())
1991 break;
1992 }
1993
1994 check_trigger_file();
1995
1996 timeout = min(100u, ops->eta_msec);
1997
1998 ret = poll(pfds, nr_clients, timeout);
1999 if (ret < 0) {
2000 if (errno == EINTR)
2001 continue;
2002 log_err("fio: poll clients: %s\n", strerror(errno));
2003 break;
2004 } else if (!ret)
2005 continue;
2006 } while (ret <= 0);
2007
2008 for (i = 0; i < nr_clients; i++) {
2009 if (!(pfds[i].revents & POLLIN))
2010 continue;
2011
2012 client = find_client_by_fd(pfds[i].fd);
2013 if (!client) {
2014 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2015 continue;
2016 }
2017 if (!fio_handle_client(client)) {
2018 log_info("client: host=%s disconnected\n",
2019 client->hostname);
2020 remove_client(client);
2021 retval = 1;
2022 } else if (client->error)
2023 retval = 1;
2024 fio_put_client(client);
2025 }
2026 }
2027
2028 fio_client_json_fini();
2029
2030 free(pfds);
2031 return retval || error_clients;
2032}